A proportional risk model for time-to-event analysis in randomized controlled trials

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Parametric models dominate regression for most outcomes, but time-to-event data predominantly uses non-parametric Cox models.
• Existing effect measures like hazard ratios and odds ratios for time-to-event data have reported disadvantages.
• The limitations of current methods hinder clear interpretation and technical application.

Purpose of the Study:

• To propose a novel parametric proportional risk model for analyzing time-to-event outcomes.
• To address interpretational and technical issues associated with current hazard and odds ratios.
• To enable calculation of absolute effect measures and enhance communication of results.

Main Methods:

• Development of a parametric proportional risk model for two-group time-to-event situations.
• Explicitly modeling risk, rather than hazard or odds, to improve interpretability.
• Parameter estimation using maximum likelihood with proper handling of censoring.

Main Results:

• The proposed model provides a more interpretable alternative to existing methods for time-to-event analysis.
• It allows for the computation of absolute effect measures, such as risk differences and numbers needed to treat.
• Results can be presented on the original time scale (accelerated/prolongated failure time), aiding non-technical understanding.

Conclusions:

• The parametric proportional risk model offers a robust and interpretable framework for time-to-event data analysis.
• It overcomes limitations of traditional hazard and odds ratios, providing absolute measures and enhanced communication.
• The model is implementable in standard statistical software and demonstrated with a diabetes drug trial example.